|
Abstract
|
Customer churn prediction is often formulated as a binary classification task. Feature selection is a significant preprocessing step that can improve the performance of the resulted churn prediction model. Its principal objective is to find a minimum set that eliminates irrelevant or redundant features and increases the performance of learning techniques. This study proposes a new feature selection method that exploits the Imperialist Competitive Algorithm to select the optimal feature set. To evaluate the usefulness of the proposed method, three state-of-the-art filter feature selection methods are selected. Also, we develop a wrapper feature selection method that works based on the genetic algorithm. We conduct the experiments using two churn datasets of the telecommunication industry. The experiments show that the proposed feature selection method considerably improves the performance of the generated models.
|